infer: A Bayesian Inference Approach towards Energy Efficient Data Collection in Dense Sensor Networks

In this paper, we propose a novel approach for efficiently sensing a remote field using wireless sensor networks. Our approach, the infer algorithm, is fully distributed, has low overhead and saves considerable energy compared to using just the data aggregation communication paradigm. This is accomplished by using a distributed algorithm to put nodes into sleep mode for a given period of time, thereby trading off energy usage for the accuracy of the data received at the sink. Bayesian inference is used to infer the missing data from the nodes that were not active during each sensing epoch. As opposed to other methods that have been considered, such as wavelet compression and distributed source coding, our algorithm has lower overhead in terms of both inter-node communication and computational complexity. Our simulations show that on average our algorithm produces energy savings of 59% while still maintaining data that is accurate to within 7.9%. We also show how the parameters of the algorithm may be tuned to optimize network lifetime for a desired level of data accuracy.